http://www.theguardian.com/technology/2 ... autonomous

I'd imagine that if you lean out the window, for instance, the car will stop. Google and the like would rather not be sued.

On the bomb front, you could say the same thing about personal drones. And considering the Americans lost one to Iranian hacking...

Wired has another take on autonomous vehicles and the harm they could cause.


http://www.wired.com/2014/05/the-robot- ... o-hit-you/

I find that a bit of a nonsensical example tbh. Surely the point of a computer controlled car is it would never get in the position where it would be responsible for a crash. It should never have any of the issues that cause humans to crash i.e. inattentiveness, incorrect decision making etc. It should only ever crash due to mechanical failure (in which case the 'driver' can't do anything) or when it's the passive partner in a collision (in which case the 'driver' can't do anything). A car driven by a less.. fallible driver than a human one should never be faced with that situation. and I say this as someone who has had two crashes (thankfully both minor and nobody hurt) that were definitely my fault.

Note : I'm entirely aware that computer systems are often less than infallible, I'm just assuming a driverless car will be really, really well tested before it gets to anything like widespread acceptancy.

Jon

If all vehicles on the road are "driverless" or computer-controlled, you'd be correct. But the problem is that as soon as you put a human driver on the road, you introduce risks, doubt etc. A computer cannot predict what human drivers will do. Too many variables. What it can do is predict likely outcomes and plan for those eg it's being tailgated by a human driver. The AI-driver can slow the car to reduce impact/damage in the event of a collision. What it can't do is avoid being rammed by the tailgater.

Indeed, but I'd put that under the 'things the computer driver can't do anything about' category. It can't help if someone else hits it.The situation suggested was where the computer driver was inevitably going to crash, and had to choose which of the two vehicles it was going to hit. My take is that would realistically never happen in any situation where the computer driver actually had any choice about it.

The computer driver has reactions far beyond any human and probably better 'spacial awareness' because it will have a 360 degree sensors that are much better than a rear view mirror - RADAR/LIDAR etc - plus better data about how fast the car is going and it will never do things like not leave enough braking room. It's just not going to be that careless. It may well be the self-driving car will get crashed into then pushed into another car and maybe it would have to decide whether to minimise damage to itself or total damage to both but I can't believe the scenario suggested actually happening in the real world.

Jon

I think it depends on what the scenario is that leads to "Sophie's choice" or whatever you'll call the decision. If on a single carriageway on a country lane and the AI-car encounters two vehicles, one overtaking the other, then the safest thing would be to avoid a collision with a human and chuck the car into a ditch.

At least until they invent turbo boost.

Sadly, not all self-driving cars will be KITT.

I know. Until then, I'll stick with my black Golf and it's red grille stripe.

I think the one that is more likely is the pedestrian walking into the road in front of the car

Does the car (assuming it cant stop / swerve to miss)


- Protect the Driver & Passengers by hitting the pedestrian
- Protect the pedestrian by swerve into the path of the nearby wall / oncoming car and possibly killing the occupants of the car

I'm not a driver, but that sounds like just as difficult a decision for a meat-pilot as for his electronic counterpart. The silicon version can hit the brakes faster than you can though, so the pedestrian will at least be hit more slowly than he otherwise would have been.

I guess the big problem in the case of a crash is liability. At the moment, we're quite happy to assume that auto-piloted cars are going to be faultless, we do have to remember that they are going to be programmed by humans. So in the event of a crash, who's going to be liable if the auto-car is at fault.

Will it be the owner/driver? If they are not controlling the car, then how can they be held responsible for a des soon made by software? Will it be the manufacturer? There'll need to be software audits, as well as black box investigations. Will a manufacturer allow their software to be investigated in a forensic matter?

I also ponder the idea of "artificial obsolescence" - we see this in many devices, where software decrees whether a physical product can use it or not. If car A can be upgraded with software with more safeguards but car B cannot simply because it's older, who is repos sidle in that situation?

What happens if legislation requires all auto-pilot vehicles to have a set of features, but the manufacturer no longer supports upgrades for that model?

All this assumes a similar business model to those adopted by phones and home computers - I expect we'll need more robust controls to ensure that all auto-cars can be updated.

Why buy an auto-driving car? Cars are depreciating assets, we only buy them because that's kind of the way we've always done things. Why not rent them, on say a six month contract? Why bother with all the MOTs and maintenance and upgrades and what have you? Why on earth would you want to own a car you're not even driving?

I suspect auto-driving cars won't be owned but the likes of you and me. It will probably be like the Boris Bike scheme in London. Cars will be pooled, you hop into the nearest one and go.

Equally, self-driving vehicles will never likely be commonplace on the roads all the while the rest of us keep wanting to drive ourselves. The technology is clever, but will only truly be safe and convenient when all vehicles are self-driving.

If you fire up an app on your phone and get a self driving car on demand wherever you are and however you got there, all for very little money because the car itself is part of a fleet in constant use, then self driving cars will be the only ones on the road before very long. That's what happens when you make life simpler and cheaper for people.

Living far outside of any city would presumably be the main reason for owning a car in such a future. Consider it one more thing that makes country living economically marginal.

Why then buy a none auto-car – it has exactly the same negatives as an auto car

The reason is convenience/cost. I could hire a car or a pfp lease and change it every 3 years but I bought one. Yes its a decreasing asset BUT its something I own and in the event of a job loss etc I will have the means of getting from A-B and finding a new job. If i was leasing I would still have to find the £x hundred / month

As to why buy an auto-driving car – I would if the price was similar and the law / technology was in sync. For example I could read in it, go to be pub and have a few beers and not worry about drink driving. Make an RV auto-drive and I travel to Scotland overnight sleeping the journey away.